Sequences of Functions
9.1 Pointwise Convergence
We have accumulated much experience working with sequences of numbers. The
next level of complexity is sequences of functions. This chapter explores several
ways that sequences of functions can converge to another function. The basic
starting point is contained in the following denitions.
Denition 9.1.1. Suppose S R and for each n N there is a function
f
n
: S R. The collection {f
n
: n N} is a sequence of functions dened on S.
For each xed x S, f
n
(x) is a sequence of numbers, and it makes sense
to ask whether this sequence converges. If f
n
(x) converges for each x S, this
denes a new function f : S R by
f(x) = lim
n
f
n
(x).
The function f is called the pointwise limit of the sequence f
n
, or, equivalently,
it is said f
n
converges pointwise to f. This is abbreviated f
n
S
f, or simply
f
n
f, if the domain is clear from the context.
Example 9.1.1. Let
f
n
(x) =
_
_
0, x 0
x
n
, 0 < x < 1
1, x 1
.
Then f
n
f where
f(x) =
_
0, x 1
1, x 1
.
(See Figure 9.1.) This example shows that a pointwise limit of continuous
functions need not be continuous.
91
92 CHAPTER 9. SEQUENCES OF FUNCTIONS
0.2 0.4 0.6 0.8 1.0
0.2
0.4
0.6
0.8
1.0
Figure 9.1: The rst ten functions from the sequence of Example 9.1.1.
3 2 1 1 2 3
0.4
0.2
0.2
0.4
Figure 9.2: The rst four functions from the sequence of Example 9.1.2.
Example 9.1.2. For each n N, dene f
n
: R R by
f
n
(x) =
nx
1 +n
2
x
2
.
(See Figure 9.2.) Clearly, each f
n
is an odd function and lim
x
f
n
(x) = 0. A
bit of calculus shows that f
n
(1/n) = 1/2 and f
n
(1/n) = 1/2 are the extreme
values of f
n
. Finally, if x = 0,
f
n
(x) =
nx
1 +n
2
x
2
<
nx
n
2
x
2
1
nx
implies f
n
0. This example shows that functions can remain bounded away
from 0 and still converge pointwise to 0.
May 1, 2009 http://math.louisville.edu/lee/ira
Pointwise and Uniform Convergence 93
0.2 0.4 0.6 0.8 1.0
10
20
30
40
50
60
Figure 9.3: The rst four functions from the sequence of Example 9.1.3.
Example 9.1.3. Dene f
n
: R R by
f(x) =
_
_
2
2n+4
x 2
n+3
,
1
2
n+1
< x <
3
2
n+2
2
2n+4
x + 2
n+4
,
3
2
n+2
x <
1
2
n
0, otherwise
To gure out what this looks like, it might help to look at Figure 9.3.
The graph of f
n
is a piecewise linear function supported on [1/2
n+1
, 1/2
n
]
and the area under the isoceles triangle of the graph over this interval is 1.
Therefore,
_
1
0
f
n
= 1 for all n.
If x > 0, then whenever x > 1/2
n
, we have f
n
(x) = 0. From this it easily
follows that f
n
0.
The lesson to be learned from this example is that it may not be true that
lim
n
_
1
0
f
n
=
_
1
0
lim
n
f
n
.
Example 9.1.4. Dene f
n
: R R by
f
n
(x) =
_
n
2
x
2
+
1
2n
, x
1
n
x, x >
1
n
.
(See Figure 9.4.) The parabolic section in the center was chosen so f
n
(1/n) =
1/n and f
n
(1/n) = 1. This splices the sections together at (1/n, 1/n) so
f
n
is dierentiable everywhere. Its clear f
n
x, which is not dierentiable
at 0.
This example shows that the limit of dierentiable functions need not be
dierentiable.
The examples given above show that continuity, integrability and dieren
tiability are not preserved in the pointwise limit of a sequence of functions. To
have any hope of preserving these properties, a stronger form of convergence is
needed.
May 1, 2009 http://math.louisville.edu/lee/ira
94 CHAPTER 9. SEQUENCES OF FUNCTIONS
1.0 0.5 0.5 1.0
0.2
0.4
0.6
0.8
1.0
Figure 9.4: The rst ten functions f from Example 9.1.4.
9.2 Uniform Convergence
Denition 9.2.1. The sequence f
n
: S R converges uniformly to f : S R
on S, if for each > 0 there is an N N so that whenever n N and x S,
then f
n
(x) f(x) < .
In this case, we write f
n
S
f, or simply f
n
f, then f
n
S
f.
May 1, 2009 http://math.louisville.edu/lee/ira
9.3. METRIC PROPERTIES OF UNIFORM CONVERGENCE 95
Proof. Let x
0
S and > 0. There is an N N such that when n N,
then f(x) f
n
(x) < for all x S. In particular, f(x
0
) f
n
(x
0
) < when
n N. This shows f
n
(x
0
) f(x
0
). Since x
0
S is arbitrary, it follows that
f
n
f.
The rst three examples given above show the converse to Theorem 9.2.1
is false. There is, however, one interesting and useful case in which a partial
converse is true.
Denition 9.2.2. If f
n
S
f and f
n
(x) f(x) for all x S, then f
n
increases
to f on S. If f
n
S
f and f
n
(x) f(x) for all x S, then f
n
decreases to f on
S. In either case, f
n
is said to converge to f monotonically.
The functions of Example 9.1.4 decrease to x. Notice that in this case,
the convergence is also happens to be uniform. The following theorem shows
Example 9.1.4 to be an instance of a more general phenomenon.
Theorem 9.2.2 (Dinis Theorem). If
(a) S is compact,
(b) f
n
S
f monotonically,
(c) f
n
C(S) for all n N, and
(d) f C(S),
then f
n
f.
Proof. There is no loss of generality in assuming f
n
f, for otherwise we con
sider f
n
and f. With this assumption, if g
n
= f
n
f, then g
n
is a sequence
of continuous functions decreasing to 0. It suces to show g
n
0.
To do so, let > 0. Using continuity and pointwise convergence, for each
x S nd an open set G
x
containing x and an N
x
N such that g
Nx
(y) <
for all y G
x
. Notice that the monotonicity condition guarantees g
n
(y) < for
every y G
x
and n N
x
.
The collection {G
x
: x S} is an open cover for S, so it must contain a nite
subcover {G
xi
: 1 i n}. Let N = max{N
xi
: 1 i n} and choose m N.
If x S, then x G
xi
for some i, and 0 g
m
(x) g
N
(x) g
Ni
(x) < . It
follows that g
n
0.
9.3 Metric Properties of Uniform Convergence
If S R, let B(S) = {f : S R : f is bounded}. For f B(S), dene
f
S
= lub{f(x) : x S}. (It is abbreviated to f, if the domain S is clear
from the context.) Apparently, f 0, f = 0 f 0 and, if g B(S),
May 1, 2009 http://math.louisville.edu/lee/ira
96 CHAPTER 9. SEQUENCES OF FUNCTIONS
then f g = g f. Moreover, if h B(S), then
f g = lub{f(x) g(x) : x S}
lub{f(x) h(x) + h(x) g(x) : x S}
lub{f(x) h(x) : x S} + lub{h(x) g(x) : x S}
= f h +h g
Combining all this, it follows that f g is a metric
1
on B(S).
The denition of uniform convergence implies that for a sequence of bounded
functions f
n
: S R,
f
n
f f
n
f 0.
Because of this, the metric f g is often called the uniform metric or the sup
metric. Many ideas developed using the metric properties of R can be carried
over into this setting. In particular, there is a Cauchy criterion for uniform
convergence.
Denition 9.3.1. Let S R. A sequence of functions f
n
: S R is a Cauchy
sequence under the uniform metric, if given > 0, there is an N N such that
when m, n N, then f
n
f
m
< .
Theorem 9.3.1. Let f
n
B(S). There is a function f B(S) such that
f
n
f i f
n
is a Cauchy sequence in B(S).
Proof. () Let f
n
f.
A collection of functions S is said to be complete under uniform convergence,
if every Cauchy sequence in S converges to a function in S. Theorem 9.3.1 shows
B(S) is complete under uniform convergence. Well see several other collections
of functions that are complete under uniform convergence.
1
Denition 2.2.3
May 1, 2009 http://math.louisville.edu/lee/ira
9.4. SERIES OF FUNCTIONS 97
9.4 Series of Functions
The denitions of pointwise and uniform convergence are extended in the natural
way to series of functions. If
k=1
f
k
is a series of functions dened on a
set S, then the series converges pointwise or uniformly, depending on whether
the sequence of partial sums, s
n
=
n
k=1
f
k
converges pointwise or uniformly,
respectively. It is absolutely convergent or absolutely uniformly convergent, if
n=1
f
n
 is convergent or uniformly convergent on S, respectively.
The following theorem is obvious and its proof is left to the reader.
Theorem 9.4.1. Let
n=1
f
n
be a series of functions dened on S. If
n=1
f
n
is absolutely convergent, then it is convergent. If
n=1
f
n
is absolutely uni
formly convergent, then it is uniformly convergent.
The following theorem is a restatement of Theorem 9.2.2 for series.
Theorem 9.4.2. If
n=1
f
n
is a series of nonnegative continuous functions
converging pointwise to a continuous function on a compact set S, then
n=1
f
n
converges uniformly on S.
A simple, but powerful technique for showing uniform convergence of series
is the following.
Theorem 9.4.3 (Weierstrass MTest). If f
n
: S R is a sequence of functions
and M
n
is a sequence nonnegative numbers such that f
n
S
M
n
for all n N
and
n=1
M
n
converges, then
n=1
f
n
is absolutely uniformly convergent.
Proof. Let > 0 and s
n
be the sequence of partial sums of
n=1
f
n
. There is
an N N such that when n > m N, then
n
k=m
M
k
< . So,
s
n
s
m
_
_
_
_
_
n
k=m+1
f
k

_
_
_
_
_
k=m+1
f
k
n
k=m
M
k
< .
This shows s
n
is a Cauchy sequence in B(S) and must converge according to
Theorem 9.3.1.
9.5 Continuity and Uniform Convergence
Theorem 9.5.1. If f
n
: S R such that each f
n
is continuous at x
0
and
f
n
S
f, then f is continuous at x
0
.
Proof. Let > 0. Since f
n
f.
To prove this theorem, we rst need a lemma.
Lemma 9.5.4. For n N let c
n
=
_
_
1
1
(1 t
2
)
n
dt
_
1
and
k
n
(t) =
_
c
n
(1 t
2
)
n
, t 1
0, t > 1
.
(See Figure 9.6.) Then
(a) k
n
(t) 0 on [1, 1] for all n N;
(b)
_
1
1
k
n
= 1 for all n N; and,
(c) if 0 < < 1, then k
n
0 on [1, ] [, 1].
1.0 0.5 0.5 1.0
0.2
0.4
0.6
0.8
1.0
1.2
Figure 9.6: Here are the graphs of k
n
(t) for n = 1, 2, 3, 4, 5.
May 1, 2009 http://math.louisville.edu/lee/ira
9.5. CONTINUITY AND UNIFORM CONVERGENCE 99
Proof. Parts (a) and (b) follow easily from the denition of k
n
.
To prove (c) rst note that
1 =
_
1
1
k
n
_
1/
n
1/
n
c
n
(1 t
2
)
n
dt c
n
2
n
_
1
1
n
_
n
.
Since
_
1
1
n
_
n
1
e
, it follow that there is an > 0 such that c
n
<
n.
2
Letting (0, 1) and t 1,
k
n
(t) k
n
()
n(1
2
)
n
0
by LHospitals Rule. Since k
n
is an even function, this establishes (c).
A sequence of functions satisfying conditions such as those in Lemma 9.5.4
is called a convolution kernel or a Dirac sequence. Several such kernels play a
key role in the study of Fourier series.
We now turn to the proof of the theorem.
Proof. There is no generality lost in assuming [a, b] = [0, 1], for otherwise we
consider the linear change of variables g(x) = f((b a)x + a). Similarly, we
can assume f(0) = f(1) = 0, for otherwise we consider g(x) = f(x) ((f(1)
f(0))x+f(0), which is a polynomial added to f. We can further assume f(x) = 0
when x / [0, 1].
Set
p
n
(x) =
_
1
1
f(x +t)k
n
(x) dt. (9.1)
To see p
n
is a polynomial, change variables in the integral using u = x + t to
arrive at
p
n
(x) =
_
x+1
x1
f(u)k
n
(u x) du =
_
1
0
f(u)k
n
(x u) du,
because f(x) = 0 when x / [0, 1]. Notice that k
n
(x u) is a polynomial in u
with coecients being polynomials in x, so integrating f(u)k
n
(x u) yields a
polynomial in x. (Just try it for a small value of n and a simple function f!)
2
It is interesting to note that with a bit of work using complex variables one can prove
cn =
(n + 3/2)
(n + 1)
=
n + 1/2
n
n 1/2
n 1
n 3/2
n 2
3/2
1
.
May 1, 2009 http://math.louisville.edu/lee/ira
910 CHAPTER 9. SEQUENCES OF FUNCTIONS
Use (9.1) and Lemma 9.5.4(b) to see for (0, 1) that
p
n
(x) f(x) =
_
1
1
f(x +t)k
n
(t) dt f(x)
_
1
1
(f(x +t) f(x))k
n
(t) dt
_
1
1
f(x +t) f(x)k
n
(t) dt
=
_
k
n
(t) dt <
2
(9.3)
According to Lemma 9.5.4(c), there is an N N so that when n N and
t , then k
n
(t) <
8(f+1)(1)
. Using this, it follows that
_
<t1
f(x +t) f(x)k
n
(t) dt
=
_
1
f(x +t) f(x)k
n
(t) dt +
_
1
k
n
(t) dt
< 2f
8(f + 1)(1 )
(1 ) + 2f
8(f + 1)(1 )
(1 ) =
2
(9.4)
Combining (9.3) and (9.4), it follows from (9.2) that p
n
(x) f(x) < for all
x [0, 1] and p
n
f.
Corollary 9.5.5. If f C([a, b]) and > 0, then there is a polynomial p such
that f p
[a,b]
< .
The theorems of this section can also be used to construct some striking
examples of functions with unwelcome behavior. Following is perhaps the most
famous.
Example 9.5.1. There is a continuous f : R R that is dierentiable nowhere.
May 1, 2009 http://math.louisville.edu/lee/ira
9.5. CONTINUITY AND UNIFORM CONVERGENCE 911
0.5 1.0 1.5 2.0 2.5 3.0 3.5
0.2
0.4
0.6
0.8
1.0
Figure 9.7: s
0
, s
1
and s
2
from Example 9.5.1.
0.5 1.0 1.5 2.0
0.5
1.0
1.5
2.0
2.5
3.0
Figure 9.8: The nowhere dierentiable function f from Example 9.5.1.
Proof. Thinking of the canonical example of a continuous function that fails to
be dierentiable at a pointthe absolute value functionwe start with a saw
tooth function.
s
0
(x) =
_
x 2n, 2n x < 2n + 1, n Z
2n + 2 x, 2n + 1 x < 2n + 2, n Z
Notice that s
0
is continuous and periodic with period 2 and maximum value 1.
Compress it both vertically and horizontally:
s
n
(x) =
_
3
4
_
n
s
n
(4
n
x) , n N.
Each s
n
is continuous and periodic with period p
n
= 2/4
n
and s
n
= (3/4)
n
.
Finally, the desired function is
f(x) =
n=0
s
n
(x).
Since s
n
= (3/4)
n
, the Weierstrass Mtest implies the series dening f is
uniformly convergent and Corollary 9.5.2 shows f is continuous on R. We will
show f is dierentiable nowhere.
May 1, 2009 http://math.louisville.edu/lee/ira
912 CHAPTER 9. SEQUENCES OF FUNCTIONS
Let x R, m N and h
m
= 1/(2 4
m
).
If n > m, then h
m
/p
n
= 4
nm1
N, so s
n
(x h
m
) s
n
(x) = 0 and
f(x h
m
) f(x)
h
m
=
m
k=0
s
k
(x h
m
) s
k
(x)
h
m
. (9.5)
On the other hand, if n < m, then a worstcase estimate is that
s
n
(x h
m
) s
n
(x)
h
m
_
3
4
_
n
/
_
1
4
n
_
= 3
n
.
This gives
m1
k=0
s
k
(x h
m
) s
k
(x)
h
m
m1
k=0
s
k
(x h
m
) s
k
(x)
h
m
3
m
1
3 1
<
3
m
2
. (9.6)
Since s
m
is linear on intervals of length 4
m
= 2 h
m
with slope 3
m
on
those linear segments, at least one of the following is true:
s
m
(x +h
m
) s(x)
h
m
= 3
m
or
s
m
(x h
m
) s(x)
h
m
= 3
m
. (9.7)
Suppose the rst of these is true. The argument is essentially the same in the
second case.
Using (9.5), (9.6) and (9.7), the following estimate ensues
f(x +h
m
) f(x)
h
m
k=0
s
k
(x +h
m
) s
k
(x)
h
m
k=0
s
k
(x +h
m
) s
k
(x)
h
m
s
m
(x +h
m
) s(x)
h
m
m1
k=0
s
k
(x h
m
) s
k
(x)
h
m
> 3
m
3
m
2
=
3
m
2
.
Since 3
m
/2 , it is apparent f
k
) R(f
N
, P, x
k
) = 
n
k=1
f(x
k
)I
k

n
k=1
f
N
(x
k
)I
k
 (9.8)
= 
N
k=1
(f(x
k
) f
N
(x
k
))I
k

k=1
f(x
k
) f
N
(x
k
)I
k

<
2(b a))
n
k=1
I
k

=
3
According to Theorem 8.5.1, there is a P P([a, b]) such that whenever
P Q
1
and P Q
2
, then
R(f
N
, Q
1
, x
k
) R(f
N
, Q
2
, y
k
) <
3
. (9.9)
Combining (9.8) and (9.9) yields
R(f, Q
1
, x
k
) R(f, Q
2
, y
k
)
= R(f, Q
1
, x
k
) R(f
N
, Q
1
, x
k
) +R(f
N
, Q
1
, x
k
)
R(f
N
, Q
1
, x
k
) +R(f
N
, Q
2
, y
k
) R(f, Q
2
, y
k
)
R(f, Q
1
, x
k
) R(f
N
, Q
1
, x
k
) + R(f
N
, Q
1
, x
k
) R(f
N
, Q
1
, x
k
)
+ R(f
N
, Q
2
, y
k
) R(f, Q
2
, y
k
)
<
3
+
3
+
3
=
Another application of Theorem 8.5.1 shows that f is integrable.
Finally,
_
b
a
f
_
b
a
f
N
_
b
a
(f f
N
)
<
_
b
a
3(b a)
=
3
May 1, 2009 http://math.louisville.edu/lee/ira
914 CHAPTER 9. SEQUENCES OF FUNCTIONS
shows that
_
b
a
f
n
_
b
a
f.
Corollary 9.6.2. If
n=1
f
n
is a series of integrable functions converging uni
formly on [a, b], then
_
b
a
n=1
f
n
=
n=1
_
b
a
f
n
Combining Theorem 9.6.1 with Dinis Theorem, gives the following.
Corollary 9.6.3. If f
n
is a sequence of continuous functions converging mono
tonically to a continuous function f on [a, b], then
_
b
a
f
n
_
b
a
f.
9.7 Uniform Convergence and Dierentiation
The relationship between uniform convergence and dierentiation is somewhat
more complex than those weve already examined. First, because there are two
sequences involved, f
n
and f
n
, either of which may converge or diverge at a
point; and second, because dierentiation is more delicate than continuity or
integration.
Example 9.1.4 is an explicit example of a sequence of dierentiable functions
converging uniformly to a function which is not dierentiable at a point. The
derivatives of the functions from that example converge pointwise to a function
that is not a derivative. The Weierstrass Approximation Theorem and Example
9.5.1 push this to the extreme by showing the existence of a sequence of poly
nomials converging uniformly to a continuous nowhere dierentiable function.
The following theorem starts to shed some light on the situation.
Theorem 9.7.1. If f
n
is a sequence of derivatives dened on [a, b] and f
n
n
= f
n
and F
n
(a) = 0, then F
n
F where F
= f.
Proof. For each n, let F
n
be an antiderivative of f
n
. By considering F
n
(x)
F
n
(a), if necessary, there is no generality lost with the assumption that F
n
(a) =
0.
Let > 0. There is an N N such that
m, n N = f
m
f
n
<
b a
.
If x [a, b] and m, n N, then the Mean Value Theorem and the assumption
that F
m
(a) = F
n
(a) = 0 yield a c [a, b] such that
F
m
(x) F
n
(x) = (F
m
(x) F
n
(x)) (F
m
(a) F
n
(a))
= f
m
(c) f
n
(c) x a f
m
f
n
(b a) < .
This shows F
n
is a Cauchy sequence in C([a, b]) and there is an F C([a, b])
with F
n
F.
It suces to show F
F
n
(x) F
n
(y)
x y
F
m
(x) F
m
(y)
x y
=
1
x y
(F
n
(x) F
m
(x)) (F
n
(y) F
m
(y))
=
1
x y
f
n
(c) f
m
(c) x y f
n
f
m
<
3
.
Letting m , it follows that
F
n
(x) F
n
(y)
x y
F(x) F(y)
x y
3
, n M. (9.11)
Fix n M and x [a, b]. Since F
n
(x) = f
n
(x), there is a > 0 so that
F
n
(x) F
n
(y)
x y
f
n
(x)
<
3
, y (x , x +) \ {x}. (9.12)
Finally, using (9.11), (9.12) and (9.10), we see
F(x) F(y)
x y
f(x)
F(x) F(y)
x y
F
n
(x) F
n
(y)
x y
+
F
n
(x) F
n
(y)
x y
f
n
(x) +f
n
(x) f(x)
F(x) F(y)
x y
F
n
(x) F
n
(y)
x y
F
n
(x) F
n
(y)
x y
f
n
(x)
+ f
n
(x) f(x)
<
3
+
3
+
3
= .
This establishes that
lim
yx
F(x) F(y)
x y
= f(x),
as desired.
May 1, 2009 http://math.louisville.edu/lee/ira
916 CHAPTER 9. SEQUENCES OF FUNCTIONS
Corollary 9.7.2. If G
n
C([a, b]) is a sequence such that G
g and
G
n
(x
0
) converges for some x
0
[a, b], then G
n
G where G
= g.
Proof. Let F
n
and F be as in Theorem 9.7.2. Then F
= g
n
g
n
= 0, so
G
n
= F
n
+
n
for some constant
n
. In particular,
n
= G
n
(x
0
) F
n
(x
0
) ,
for some R, because both G
n
and F
n
converge at x
0
.
Let > 0. Since both F
n
and
n
are Cauchy sequences, there exists an
N N such that
m, n N = F
n
F
m
<
2
and 
n
m
 <
2
.
If m, n N and x [a, b], then
G
n
(x) G
m
(x) = (F
n
(x) +
n
) (F
m
(x) +
m
)
F
n
(x) F
m
(x) + 
n
m

< F
n
F
m
+
2
< .
This shows G
n
is a Cauchy sequence in C([a, b]) and therefore G
n
G. It is
clear that G = F +, so G
= F
= g.
Corollary 9.7.3. If f
n
is a sequence of dierentiable functions dened on [a, b]
such that
k=1
f(x
0
) exists for some x
0
[a, b] and
k=1
f
n
converges uni
formly, then
_
k=1
f
_
k=1
f
n=0
a
n
(x c)
n
. (9.13)
The domain of f is the set of all x at which the series converges. The constant
c is called the center of the series.
To determine the domain of (9.13), let x R \ {c} and use the root test to
see the series converges when
limsupa
n
(x c)
n

1/n
= x c limsupa
n

1/n
< 1
May 1, 2009 http://math.louisville.edu/lee/ira
9.8. POWER SERIES 917
and diverges when
x c limsupa
n

1/n
> 1.
If r limsupa
n

1/n
1 for some r 0, then these inequalities imply (9.13) is
absolutely convergent when x c < r. In other words, if
R = lub{r : r limsupa
n

1/n
< 1}, (9.14)
then the domain of (9.13) is an interval of radius R centered at c. The root test
gives no information about convergence when x c = R. This R is called the
radius of convergence of the power series. Assuming R > 0, the open interval
centered at c with radius R is called the interval of convergence. It may be
dierent from the domain of the series because the series may converge at one
endpoint or both endpoints of the interval of convergence.
The ratio test can also be used to determine the radius of convergence, but,
as shown in (4.2), it will not work as often as the root test. When it does,
R = lub{r : r limsup
a
n+1
a
n
n=0
x
n
has center 0, radius of convergence 1 and domain (1, 1).
Example 9.8.2. For the power series
n=1
2
n
(x + 2)
n
/n, we compute
limsup
_
2
n
n
_
1/n
= 2 = R =
1
2
.
Since the series diverges when x = 2
1
2
, it follows that the interval comprising
its domain is (5/2, 3/2).
Example 9.8.3. The power series
n=1
x
n
/n has interval of convergence (1, 1)
and domain [1, 1). Notice it is not absolutely convergent when x = 1.
Example 9.8.4. The power series
n=1
x
n
/n
2
has interval of convergence (1, 1),
domain [1, 1] and is absolutely convergent on its whole domain.
The preceding is summarized in the following theorem.
Theorem 9.8.1. Let the power series be as in (9.13) and R be given by either
(9.14) or (9.15).
(a) If R = 0, then the domain of the series is {c}.
(b) If R > 0 the series converges at x when c x < R and diverges at x
when c x > R.
(c) If R (0, ), then the series may converge at none, one or both of c R
and c +R.
May 1, 2009 http://math.louisville.edu/lee/ira
918 CHAPTER 9. SEQUENCES OF FUNCTIONS
9.8.2 Uniform Convergence of Power Series
The partial sums of a power series are a sequence of polynomials converging
pointwise on the domain of the series. As has been seen, pointwise convergence
is not enough to say much about the behavior of the power series. The following
theorem opens the door to a lot more.
Theorem 9.8.2. A power series converges absolutely and uniformly on compact
subsets of its interval of convergence.
Proof. There is no generality lost in assuming the series has the form of (9.13)
with c = 0. Let the radius of convergence R > 0 and K be a compact sbset of
(R, R). Choose r (lub{x : x K}, R). If x K, then a
n
x
n
 < a
n
r
n
 for
n N. Since
n=0
a
n
r
n
 converges, the Weierstrass Mtest shows
n=0
a
n
x
n
is absolutely and uniformly convergent on K.
The following two corollaries are immediate consequences of Corollary 9.5.2
and Theorem 9.6.1, respectively.
Corollary 9.8.3. A power series is continuous on its interval of convergence.
Corollary 9.8.4. If [a, b] is an interval contained in the interval of convergence
for the power series
n=0
a
n
(x c)
n
, then
_
b
a
n=0
a
n
(x c)
n
=
n=0
a
n
_
b
a
(x c)
n
.
The next question is: What about dierentiability?
Notice that the continuity of the exponential function and LHospitals Rule
give
lim
n
n
1/n
= lim
n
exp
_
lnn
n
_
= exp
_
lim
n
lnn
n
_
= exp(0) = 1.
Therefore, for any sequence a
n
,
limsup(na
n
)
1/n
= limsupn
1/n
a
1/n
n
= limsupa
1/n
n
. (9.16)
Now, suppose the power series
n=0
a
n
x
n
has a nontrivial interval of con
vergence, I. Formally dierentiating the power series termbyterm gives a new
power series
n=1
na
n
x
n1
. According to (9.16) and Theorem 9.8.1, the term
byterm dierentiated series has the same interval of convergence as the original.
Its partial sums are the derivatives of the partial sums of the original series and
Theorem 9.8.2 guarantees they converge uniformly on any compact subset of I.
Corollary 9.7.3 shows
d
dx
n=0
a
n
x
n
=
n=0
d
dx
a
n
x
n
=
n=1
na
n
x
n1
, x I.
This process can be continued inductively to obtain the same results for all
higher order derivatives. We have proved the following theorem.
May 1, 2009 http://math.louisville.edu/lee/ira
9.8. POWER SERIES 919
Theorem 9.8.5. If f(x) =
n=0
a
n
(x c)
n
is a power series with nontrivial
interval of convergence, I, then f is dierentiable to all orders on I with
f
(m)
(x) =
n=m
n!
(n m)!
a
n
(x c)
nm
. (9.17)
Moreover, the dierentiated series has I as its interval of convergence.
9.8.3 Taylor Series
Suppose f(x) =
n=0
a
n
x
n
has I = (R, R) as its interval of convergence for
some R > 0. According to Theorem 9.8.5,
f
(m)
(0) =
m!
(mm)!
a
m
= a
m
=
f
(m)
(0)
m!
, m .
Therefore,
f(x) =
n=0
f
(n)
(0)
n!
x
n
, x I.
This is a remarkable result! It shows that the values of f on I are completely
determined by its values on any neighborhood of 0. This is summarized in the
following theorem.
Theorem 9.8.6. If a power series f(x) =
n=0
a
n
(x c)
n
has nontrivial
interval of convergence I, then
f(x) =
n=0
f
(n)
(c)
n!
(x c)
n
, x I. (9.18)
The series (9.18) is called the Taylor series
3
for f centered at c. The Taylor
series can be formally dened for any function that has derivatives of all orders
at c, but, as Example 7.4.2 shows, there is no guarantee it will converge to the
function anywhere except at c.
9.8.4 The Endpoints of the Interval of Convergence
We have seen that at the endpoints of its interval of convergence a power series
may diverge or even absolutely converge. A natural question when it does
converge is the following: What is the relationship between the value at the
endpoint and the values inside the interval of convergence?
Theorem 9.8.7 (Abel). If f(x) =
n=0
a
n
(x c)
n
has a nite radius of
convergence R > 0 and f(c +R) exists, then c +R C(f).
3
When c = 0, it is often called the Maclaurin series for f.
May 1, 2009 http://math.louisville.edu/lee/ira
920 CHAPTER 9. SEQUENCES OF FUNCTIONS
Proof. It can be assumed c = 0 and R = 1. There is no loss of generality with
either of these assumptions because otherwise just replace f(x) with f((x +
c)/R). Set s = f(1), s
1
= 0 and s
n
=
n
k=0
a
k
for n . For x < 1,
n
k=0
a
k
x
k
=
n
k=0
(s
k
s
k1
)x
k
=
n
k=0
s
k
x
k
k=1
s
k1
x
k
= s
n
x
n
+
n1
k=0
s
k
x
k
x
n1
k=0
s
k
x
k
= s
n
x
n
+ (1 x)
n1
k=0
s
k
x
k
When n , since s
n
is bounded,
f(x) = (1 x)
k=0
s
k
x
k
. (9.19)
Since (1 x)
n=0
x
n
= 1, (9.19) implies
f(x) s =
(1 x)
k=0
(s
k
s)x
k
. (9.20)
Let > 0. Choose N N such that s
N
s < /2, (0, 1) so
k=0
s
n
s < /2.
and 1 < x < 1. With these choices, (9.20) becomes
f(x) s
(1 x)
N
k=0
(s
k
s)x
k
(1 x)
k=N+1
(s
k
s)x
k
<
N
k=0
s
k
s +
2
(1 x)
k=N+1
x
k
<
2
+
2
=
It has been shown that lim
x1
f(x) = f(1), so 1 C(f).
Abels theorem opens up a more general idea for the summation of series.
Suppose, as in the proof of the theorem, f(x) =
n=0
a
n
x
n
with interval of
May 1, 2009 http://math.louisville.edu/lee/ira
9.8. POWER SERIES 921
convergence (1, 1). If lim
x1
f(x) exists, then this limit is called the Abel sum
of the coecients of the series. In this case, the following notation is used
lim
x1
f(x) = A
n=0
a
n
.
From Theorem 9.8.7, it is clear that when
n=0
a
n
exists, then
n=0
a
n
=
A
n=0
a
n
. The converse of this statement may not be true.
Example 9.8.5. Notice that
n=0
(x)
n
=
1
1 +x
,
has interval of convergence (1, 1) and diverges when x = 1. (It is the sum
1 1 + 1 1 + .) But,
A
n=0
(1)
n
= lim
x1
1
1 +x
=
1
2
.
Finally, here is an example showing the power of these techniques.
Example 9.8.6. The series
n=0
(1)
n
x
2n
=
1
1 +x
2
has (1, 1) as its interval of convergence and (0, 1] as its domain. If 0 x < 1,
then Corollary 9.6.2 justies
arctan(x) =
_
x
0
dt
1 +t
2
=
_
x
0
n=0
(1)
n
t
2n
dt =
n=0
(1)
n
2n + 1
x
2n+1
.
This series for the arctangent converges by the alternating series test when
x = 1, so Theorem 9.8.7 implies
n=0
(1)
n
2n + 1
= lim
x1
arctan(x) = arctan(1) =
4
.
May 1, 2009 http://math.louisville.edu/lee/ira
922 CHAPTER 9. SEQUENCES OF FUNCTIONS
May 1, 2009 http://math.louisville.edu/lee/ira
Much more than documents.
Discover everything Scribd has to offer, including books and audiobooks from major publishers.
Cancel anytime.